Mathematical Reasoning on GSM8K (test) (Accuracy, Convergence, Variance)
92.4AccuracyChordal + PEO Repair (Ours)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Chordal + PEO Repair (Ours)Topology Architecture=Chordal + PEO Repair, Base Model=LLaMA-3-8B, Ensemble Size (N)=52026.02 | 92.4 | 100 | 14.2 | 3.1 | |
| Majority Vote (MoE)Topology Architecture=Majority Vote (MoE), Base Model=LLaMA-3-8B, Ensemble Size (N)=52026.02 | 84.5 | 100 | 1 | 0 | |
| Single Agent (Baseline)Topology Architecture=Single Agent, Base Model=LLaMA-3-8B, Ensemble Size (N)=52026.02 | 78.1 | — | 1 | — | |
| Frustrated (Unmediated)Topology Architecture=Frustrated (Unmediated), Base Model=LLaMA-3-8B, Ensemble Size (N)=52026.02 | 0 | 0 | 50 | 92.4 |